WEBVTT - The Hater's Guide To The AI Bubble, Pt. 3 0:00:02.560 --> 0:00:06.920 Zone Media, stop your grinnin and drop your linen. I'm 0:00:07.000 --> 0:00:21.200 d zechron and this is better offline. Welcome to the 0:00:21.200 --> 0:00:23.319 final part of our three part Hater's Guide to the 0:00:23.320 --> 0:00:25.599 AI Bubble, where we've been talking about the growing cracks 0:00:25.640 --> 0:00:27.960 emerging in the industry and the potential contagion to the 0:00:27.960 --> 0:00:30.840 wider economy that are pullback in general of AI spending 0:00:30.880 --> 0:00:33.480 will cause. And I must be clear, things are bad 0:00:33.640 --> 0:00:36.480 the vibes A rand said. I spent the past two episodes, 0:00:36.520 --> 0:00:38.920 and we'll spend much of this episode explaining why. But 0:00:38.960 --> 0:00:40.840 I just want to recap something from the first part 0:00:40.840 --> 0:00:44.880 of this series. Right now, seven companies, the Magnificent Seven, 0:00:45.080 --> 0:00:48.000 account for the value of nearly thirty five percent of 0:00:48.040 --> 0:00:50.320 all of US stocks, and the biggest of these companies 0:00:50.360 --> 0:00:52.440 is in Nvidia, which has gotten fat off the back 0:00:52.479 --> 0:00:55.880 of the AI bubble today a disproportionate amount of Nvidia's 0:00:55.920 --> 0:00:58.120 revenue eighty eight percent of it is coming from the 0:00:58.160 --> 0:01:00.720 other members of the Magnificent Seven. And there of the 0:01:00.840 --> 0:01:03.920 enterprise GPUs designed to power the compute heavy eight A 0:01:04.040 --> 0:01:07.199 applications that you're hearing about all day. If and when 0:01:07.280 --> 0:01:11.240 they pull back, invidiabile sink so too will the other companies, 0:01:11.319 --> 0:01:13.360 especially when it becomes evident that they've spent nearly half 0:01:13.400 --> 0:01:17.199 a trillion dollars in two years on bugger roll. It'll 0:01:17.200 --> 0:01:20.960 be an apocalyptic shedding of value. They won't die, but 0:01:21.280 --> 0:01:23.080 it'll be the likes of which we haven't seen since 0:01:23.080 --> 0:01:24.840 the dark days of two thousand and seven and two 0:01:24.840 --> 0:01:27.959 thousand and eight for those companies. I don't know about 0:01:27.959 --> 0:01:31.480 the global banking system imploding, but I also can't say 0:01:31.520 --> 0:01:33.679 what I just I don't know. But what I do 0:01:33.800 --> 0:01:36.480 know is tech is going to take such a precipitous 0:01:36.480 --> 0:01:38.679 haircut that you're going to see scissor gashes in the 0:01:38.680 --> 0:01:42.600 man's bald head. Terrible metaphor. If that sounds like a 0:01:42.600 --> 0:01:45.800 them problem, though, and not a you problem, ask yourself, 0:01:45.840 --> 0:01:47.720 what's in your fuah one K? Do you own any 0:01:47.800 --> 0:01:50.480 index funds or ETFs? Are you close to retirement? And 0:01:50.480 --> 0:01:52.800 how much do you enjoy eating ramen? Are you willing 0:01:52.840 --> 0:01:54.800 to eat the same whiskers cat food as your American 0:01:54.840 --> 0:01:57.840 short hair? I am, except I feed them a different, 0:01:57.880 --> 0:02:00.440 better food because Babu deserves the best. But in the 0:02:00.480 --> 0:02:02.520 last two episodes, I made the case that the generative 0:02:02.560 --> 0:02:05.880 AI industry is unique in its capital demands and unparalleled 0:02:05.880 --> 0:02:07.840 in the fact that no company beside in video or 0:02:07.880 --> 0:02:10.840 consultancies are making any money from them. I can easily 0:02:10.840 --> 0:02:13.800 imagine the chain of events where one big company fails, 0:02:13.880 --> 0:02:15.760 like either because they ran out of money or because 0:02:15.800 --> 0:02:18.440 the investors pulled back. This in turn spooks the rest 0:02:18.440 --> 0:02:21.600 of the market, causing other companies to start seeing massive 0:02:21.639 --> 0:02:24.040 revenue drops. And as the demand for compute drops, hyper 0:02:24.040 --> 0:02:28.000 scalers like Microsoft, Amazon and Google massively cut their capital expenditures, 0:02:28.160 --> 0:02:31.160 canceling data center build outs and of course slowing down 0:02:31.200 --> 0:02:34.160 their GPU orders. And then Jensen Huang would have to 0:02:34.160 --> 0:02:36.160 watch as the cash cow that made his company rich 0:02:36.200 --> 0:02:38.440 and fat is take over a walk behind the barn 0:02:39.280 --> 0:02:41.760 for two shots to ring out. Though I must be clear, 0:02:42.280 --> 0:02:48.600 video isn't gonna die, None of the Magnificent seven will. However, 0:02:49.000 --> 0:02:51.120 the first big failure, the first domino to fall, could 0:02:51.120 --> 0:02:52.959 be any number of companies, and while I don't think 0:02:52.960 --> 0:02:55.680 open Ai and Anthropic are necessarily going to be the first, 0:02:56.440 --> 0:02:59.040 I also believe that there's an air of inevitability about 0:02:59.080 --> 0:03:02.120 one of or both of their demises. This matter is 0:03:02.160 --> 0:03:04.560 because these companies are for all intents and purposes the 0:03:04.560 --> 0:03:08.639 generative AI industry. They're deeply unsustainable and unstable, and yet 0:03:08.680 --> 0:03:11.720 they are also critical to the AI trade. Continuing, I 0:03:11.720 --> 0:03:14.959 haven't really spent much time on my favorite subject, that 0:03:15.240 --> 0:03:17.720 open ai is a systemic risk to the tech industry. 0:03:17.880 --> 0:03:21.639 But let's recap my reporting. Open Ai and Anthropic both 0:03:21.680 --> 0:03:23.880 lose billions of dollars a year after revenue, and their 0:03:23.919 --> 0:03:26.760 stories do not mirror any other startup in history, not Uber, 0:03:26.960 --> 0:03:30.160 not Amazon Web Services, and nothing. I've addressed my Uber 0:03:30.240 --> 0:03:34.080 point before in how does open ai survive? It'll be 0:03:34.120 --> 0:03:36.560 in this show notes. Someone always asks me about this. 0:03:36.720 --> 0:03:40.200 Please stop, learn to read, learn to listen. Soft Bank 0:03:40.280 --> 0:03:42.640 also is putting itself in dire straits simply to fund 0:03:42.640 --> 0:03:45.800 open Ai. The steel threatens its credit ring, with SoftBank 0:03:45.840 --> 0:03:47.680 having to take on what will be multiple loans to 0:03:47.680 --> 0:03:50.200 fund the remaining thirty billion dollars open ai is forty 0:03:50.200 --> 0:03:52.600 billion dollar funding round which is yet to close, and 0:03:52.880 --> 0:03:55.400 open aiy is in fact still raising from the Middle East. 0:03:56.240 --> 0:03:58.680 This is before you consider the other nineteen billion dollars 0:03:58.720 --> 0:04:00.960 that soft Bank has agreed to contribute to the Stargate 0:04:01.040 --> 0:04:03.920 Data Center project, money that it does not have available 0:04:03.960 --> 0:04:07.440 at this time. But I must say, while recording this episode, 0:04:07.640 --> 0:04:09.280 I saw some news come out and I'm just going 0:04:09.320 --> 0:04:11.920 to read it to you and said soft Bank and 0:04:11.960 --> 0:04:15.720 open Ai is five hundred billion AI project struggles to 0:04:15.720 --> 0:04:18.880 get off ground. The Stargate venture introduced a white house event, 0:04:19.000 --> 0:04:21.440 is now setting the more modest goal of building a 0:04:21.480 --> 0:04:24.719 small data center by year. And I just want to say, 0:04:25.480 --> 0:04:29.200 anyone that doubted me can suck one. I was oinking 0:04:29.240 --> 0:04:32.880 and squawking around my room just before this when I 0:04:32.920 --> 0:04:34.440 got this news in. I just want to say, I 0:04:34.600 --> 0:04:36.720 just learned this news and I'm still taking it in. 0:04:37.000 --> 0:04:39.080 I'm still taking in because it makes me so excited 0:04:39.120 --> 0:04:41.840 and sad. Anyway, neither soft Bank nor open Ai has 0:04:41.880 --> 0:04:44.840 the money for Stargate, regardless of what they built. Also, 0:04:45.160 --> 0:04:48.040 Clammy Samy's making some interesting claims about what Stargate is. 0:04:48.080 --> 0:04:51.479 He's claimed the Denton site, the core Weavers building is 0:04:51.520 --> 0:04:55.800 also Stargate. This motherfucker loves lying anyway. Open A must 0:04:55.839 --> 0:04:57.680 also convert to a for profit by the end of 0:04:57.680 --> 0:04:59.600 twenty twenty five, all whose twenty billion dollars of the 0:04:59.640 --> 0:05:02.080 remaining thirty billion dollars of funding. If it does not 0:05:02.080 --> 0:05:04.880 convert by October twenty twenty six, it's current funding converts 0:05:04.880 --> 0:05:08.280 to debt. It is also demanding remarkable and reasonable concessions 0:05:08.320 --> 0:05:11.679 and Microsoft stuff like they want to reduce their rev share, 0:05:11.920 --> 0:05:14.960 They want to reduce their their I think, their ability 0:05:15.000 --> 0:05:18.280 to their exclusive rights to sell models. I did I 0:05:18.279 --> 0:05:19.960 really need to do a monologue about this, and I 0:05:20.040 --> 0:05:23.000 know this is technically a monologue, but nevertheless, they want 0:05:23.040 --> 0:05:25.320 these insane things, and Microsoft is refusing to budge, and 0:05:25.360 --> 0:05:27.839 it's willing to walk away from negotiations necessary to convert. 0:05:27.880 --> 0:05:31.160 According to the Financial Times, and also open Ai does 0:05:31.160 --> 0:05:33.680 not have a path of profitability, and its future, like Anthropics, 0:05:33.760 --> 0:05:36.039 is dependent on the continual flow of capital from venture 0:05:36.040 --> 0:05:39.760 capitalists and big tech, who must also continue to expand infrastructure. 0:05:40.760 --> 0:05:42.440 I don't know how that's going to fucking happen with 0:05:42.480 --> 0:05:45.960 Stargate going. God, I feel so good about that. Oh 0:05:46.360 --> 0:05:49.560 I want to like bottle that up and drink it. Anyway. 0:05:50.279 --> 0:05:52.760 Anthropic is in a similar but slightly better position. It's 0:05:52.760 --> 0:05:54.760 set to lose three billion dollars this year on four 0:05:54.760 --> 0:05:57.200 billion dollars of revenue. It also has no path of 0:05:57.200 --> 0:05:59.960 profitability and recently jacked up prices on cursor, it's larger 0:06:00.120 --> 0:06:02.600 customer has had to put restraints on clawed code after 0:06:02.640 --> 0:06:05.359 allowing users to burn one hundred to ten thousand percent 0:06:05.400 --> 0:06:07.760 of their revenue a month. These are the actions of 0:06:07.760 --> 0:06:11.640 a desperate company. Nevertheless, Open Ai and Anthropics revenue amount to, 0:06:12.000 --> 0:06:14.159 by my estimates, more than half of the entire revenue 0:06:14.160 --> 0:06:17.600 of the generative AI industry, including the hyperscalers. And to 0:06:17.600 --> 0:06:20.360 be abundantly clear, the two companies that amount to a 0:06:20.440 --> 0:06:26.880 round half of all generative AI revenue are only losing money. Now. 0:06:26.880 --> 0:06:28.400 I said a lot of this before, which is why 0:06:28.440 --> 0:06:30.080 I'm trying not to harp on it too much. But 0:06:30.120 --> 0:06:32.960 the most important company in the entire AI industry needs 0:06:33.000 --> 0:06:35.960 to convert to a for profit entity from a nonprofit entity, 0:06:36.120 --> 0:06:38.120 which has never happened before in this case, by the 0:06:38.240 --> 0:06:40.400 end of the year or it's effectively dead. And even 0:06:40.400 --> 0:06:43.200 if it does, Open AI burns billions and billions and 0:06:43.240 --> 0:06:46.120 billions of dollars a year and will die without continual funding. 0:06:46.360 --> 0:06:49.200 They have no path to profitability in anyone telling you 0:06:49.240 --> 0:06:51.960 otherwise as a liar or a fantasist. And on top 0:06:51.960 --> 0:06:54.080 of this is part of the thing about SoftBank, by 0:06:54.120 --> 0:06:57.800 the Way and Stargate. They revealed that they've promised Oracle 0:06:57.880 --> 0:07:02.640 thirty billion dollars a year their cloud contract. I swear 0:07:02.640 --> 0:07:06.200 to fucking God does no one read these stories and 0:07:06.360 --> 0:07:09.960 think does this not sound just like an egregious lie? Anyway, 0:07:10.080 --> 0:07:12.080 as I wrote earlier in the year, there is really 0:07:12.080 --> 0:07:15.360 no significant adoption of generative AI services or products. Chat 0:07:15.360 --> 0:07:18.119 GPTs five hundred million weekly users, and otherwise it seems 0:07:18.120 --> 0:07:21.160 that other services they struggled to get fifteen million users. 0:07:21.520 --> 0:07:24.880 Although you could mention Google that's not a fair comparisons. 0:07:24.920 --> 0:07:28.040 It cheated by combining Google Gemini with Google Assystem to 0:07:28.080 --> 0:07:30.360 claim it had three hundred and fifty million active users 0:07:31.400 --> 0:07:33.880 a while the five hundred million weekly users sounds and 0:07:34.200 --> 0:07:36.760 is in fairness impressive. There's a world of difference between 0:07:36.800 --> 0:07:39.000 someone using your product as part of their job, which 0:07:39.000 --> 0:07:40.920 they pay for, and someone dicking around with an image 0:07:40.960 --> 0:07:43.760 generator or a college student trying to cheat on their homework. 0:07:43.960 --> 0:07:46.840 On top of that, now I'm recording this July twenty first, 0:07:46.840 --> 0:07:48.760 there's an ICO story that came out today with some 0:07:48.800 --> 0:07:52.560 bullshit vanity metrics, rope and AI. But that story said 0:07:52.560 --> 0:07:54.760 the Open AI themselves had said that they have five 0:07:54.840 --> 0:07:57.800 hundred million weekly active users. Now here's the crazy thing. 0:07:58.040 --> 0:08:01.000 That five hundred million number. That from March. That's and 0:08:01.080 --> 0:08:03.880 March now, but I came from an open AI funding announcement. 0:08:04.760 --> 0:08:08.080 Have they platowed stick around? I'm going to find out. 0:08:08.400 --> 0:08:10.640 But this is all worrying on so many levels, chief 0:08:10.640 --> 0:08:12.520 of which is that everybody has been talking about AI 0:08:12.600 --> 0:08:14.920 for three goddamn years, and everybody has said AI and 0:08:14.960 --> 0:08:17.760 every earnings and media appearance, exhausting fucking blog posts, and 0:08:17.800 --> 0:08:20.960 we still can't scrape together the bits to make functional industry. 0:08:22.160 --> 0:08:22.320 Now. 0:08:22.360 --> 0:08:23.680 I know some of you will probably read this in 0:08:23.680 --> 0:08:26.240 point to chat gipt's users, and I quote myself writing 0:08:26.240 --> 0:08:28.320 in my newsletter make fun of them, which also read 0:08:28.360 --> 0:08:31.600 for this podcast. Chat gpt is allegedly five hundred million 0:08:31.600 --> 0:08:34.280 weekly hatieth users and by the last count, only fifteen 0:08:34.320 --> 0:08:38.040 point five million paying subscribers, an absolutely futrid conversion rate, 0:08:38.080 --> 0:08:40.240 even before you realize that the actual conversion rate would 0:08:40.240 --> 0:08:43.080 be monthly active subscribers, which is how any real software 0:08:43.080 --> 0:08:46.040 company actually defines its metrics. By the fucking way, Why 0:08:46.120 --> 0:08:48.640 is this impressive? Because it grew so fast it literally 0:08:48.720 --> 0:08:51.000 had more pr and more marketing and more attention and 0:08:51.040 --> 0:08:53.840 more opportunities to sell to more people than any company 0:08:53.840 --> 0:08:56.680 has ever had in the history of anything. Every single 0:08:56.720 --> 0:08:59.000 industry has been told to think about AI for three years, 0:08:59.040 --> 0:09:00.520 and they've been told to do so because of a 0:09:00.520 --> 0:09:03.560 company called open Ai. There isn't a single goddamn product 0:09:03.559 --> 0:09:06.440 since Google or Facebook bes's had this level of media pressure, 0:09:06.640 --> 0:09:08.839 and both of these companies launched without the massive amount 0:09:08.880 --> 0:09:11.960 of media and social media that we have today. CHADGBT 0:09:12.160 --> 0:09:15.640 is a very successful growth product and an absolutely horrifying business. 0:09:16.240 --> 0:09:19.360 They're a banana republic that cannot function on their own. 0:09:19.760 --> 0:09:22.480 They do not resemble Uber, Amazon, Web Services or any 0:09:22.520 --> 0:09:24.480 other business in the past other than we work the 0:09:24.559 --> 0:09:26.920 other company that SoftBank spent way too much money on, 0:09:27.400 --> 0:09:29.800 and outside of chad GBT, there really isn't anything else. 0:09:30.880 --> 0:09:33.480 But before I wrap up this three parter and I'm tired, 0:09:33.520 --> 0:09:35.680 and I imagine you are too, I do want to 0:09:35.720 --> 0:09:39.080 address something. I want to address the kind of both 0:09:39.160 --> 0:09:41.760 sides thing that everyone's doing, and I think we all 0:09:41.800 --> 0:09:44.000 need to do way less of this because at this 0:09:44.160 --> 0:09:49.040 point it's kind of silly. Yes, Generati, if AI has functionality, 0:09:49.400 --> 0:09:51.840 there are coding products and search products that people like 0:09:51.920 --> 0:09:54.400 and some even pay for them. As I have discussed 0:09:54.400 --> 0:09:56.600 across this series and I said discuss there, but I'm 0:09:56.679 --> 0:10:00.439 keeping going and across countless podcasts and newslayers. None of 0:10:00.520 --> 0:10:03.280 these companies are profitable until one of them is profitable, 0:10:03.559 --> 0:10:07.680 really profitable sustainably, so generative AI based companies are not 0:10:07.800 --> 0:10:10.840 real businesses in any case. The problem isn't so much 0:10:10.880 --> 0:10:13.080 that l elms don't do anything, but that people talk 0:10:13.120 --> 0:10:15.760 about them doing things they can't. The use of the 0:10:15.760 --> 0:10:18.040 word agent is a deliberate attempt to suggest that l 0:10:18.040 --> 0:10:20.840 elms are autonomous, and any and all stories about AI 0:10:20.920 --> 0:10:24.560 replacing jobs are intentionally manipulative attempts to boost stock valuations 0:10:24.640 --> 0:10:27.079 and suggests that models are capable of replacing human workers 0:10:27.080 --> 0:10:30.320 at scale, which they are not. Allison Morrow of CNN, 0:10:30.320 --> 0:10:31.880 who's been on the show a few times, also has 0:10:31.880 --> 0:10:33.520 an excellent piece on this, which I've linked to in 0:10:33.559 --> 0:10:36.520 the show notes. As I discussed in my newsletter Sincerity 0:10:36.520 --> 0:10:38.800 Wins the War, which I've also linked to, this is 0:10:38.840 --> 0:10:40.800 one of the most egregious failures of the tech media 0:10:40.800 --> 0:10:44.400 I've ever seen, willingly pushing warrio Amma days outright making 0:10:44.480 --> 0:10:47.000 up of stuff. No one asked him if that if 0:10:47.000 --> 0:10:50.080 the stat he said ten to twenty percent white collar unemployment, 0:10:51.000 --> 0:10:53.079 if I started lying, I just really want to be 0:10:53.120 --> 0:10:56.200 clear about something. If I went on this podcast and 0:10:56.240 --> 0:10:59.240 I just fucking made up stats, if I was just like, yes, 0:10:59.320 --> 0:11:03.480 seventy five percent of all generative ais CEOs, literally they 0:11:03.960 --> 0:11:06.920 love kicking puppies, they love drop kicking puppies, you'd get 0:11:06.920 --> 0:11:09.360 mad at me. Right. What's the difference between that and 0:11:09.520 --> 0:11:12.920 Wario Ama Day saying twenty percent white collar and unemployment. 0:11:12.960 --> 0:11:15.480 There is none. They're both lies. And I don't know 0:11:15.520 --> 0:11:17.319 why I'm doing the dog kicking one, but it's just 0:11:17.360 --> 0:11:20.160 an example. We move on. And as a side note, 0:11:20.200 --> 0:11:24.480 the discussion also of the term AGI artificial general intelligence, 0:11:24.520 --> 0:11:27.440 and every acceptance of it is an attempt to suggest 0:11:27.480 --> 0:11:30.480 that large language models can create conscious intelligence, a fictional 0:11:30.480 --> 0:11:33.560 concept that even Meta's chief AI scientist says won't come 0:11:33.760 --> 0:11:37.960 from scaling up llms. Members of the media, listen to me, 0:11:38.760 --> 0:11:41.160 every time you talk about the really smart engineers that 0:11:41.280 --> 0:11:44.920 say Meta is paying one hundred million dollars two know 0:11:45.040 --> 0:11:47.640 that you are doing free marketing for these companies when 0:11:47.640 --> 0:11:49.840 what's really happening is people are giving tens of millions 0:11:49.880 --> 0:11:51.400 of dollars, the guys who will work on teams that 0:11:51.440 --> 0:11:54.440 are pursuing a totally unproven concept. It is like saying 0:11:54.440 --> 0:11:56.679 that they are all going to see if they could. 0:11:57.040 --> 0:11:59.000 I don't know. I was about to say the wrong 0:11:59.120 --> 0:12:02.160 create the wrong trend from Rollss and Grummet, Rollis and 0:12:02.400 --> 0:12:06.120 Wallace and Grummet. But you know those are one hundred 0:12:06.120 --> 0:12:08.679 percent possible. I'm sure someone could build them. How about this, 0:12:08.760 --> 0:12:10.320 They all believe they're going to hunt and kill the 0:12:10.360 --> 0:12:13.440 tooth Fairy because the tooth Fairy is about as realistic 0:12:13.480 --> 0:12:16.920 as agi, and if you disagree with me, emummy, I 0:12:17.000 --> 0:12:19.199 might even read him. I should also add that the 0:12:19.280 --> 0:12:22.360 use of the word singularity is similarly manipulative. These are 0:12:22.920 --> 0:12:26.000 terms that are used to obfuscate the actual abilities of 0:12:26.080 --> 0:12:29.880 large language models and indeed what can be built on them. 0:12:30.160 --> 0:12:35.880 It's really frustrating because these should be really obvious tricks, 0:12:36.920 --> 0:12:40.079 but the media keeps accepting them. And I don't want 0:12:40.120 --> 0:12:42.880 to accuse anyone of being anything. I just want to 0:12:42.880 --> 0:12:44.800 say I find it fucking disgusting. And when this all 0:12:44.840 --> 0:12:47.920 falls apart, there will be a referendum on this. I 0:12:48.000 --> 0:12:51.240 will not be kind or merciful to the people that 0:12:51.360 --> 0:12:54.640 have continually pushed these narratives. I also want to add 0:12:54.679 --> 0:12:58.199 there's another really annoying narrative that we need to get 0:12:58.320 --> 0:13:01.320 rid of, and that's these stories models lying, cheating, and 0:13:01.320 --> 0:13:04.360 stealing to reach their goals or stop themselves being turned off. 0:13:04.400 --> 0:13:07.280 These are intentionally deceptive, as these models can and clearly 0:13:07.360 --> 0:13:11.600 are being prompted to take these actions. To be abundantly clear, 0:13:11.640 --> 0:13:14.320 the manipulative suggestion here is that these models are autonomous 0:13:14.360 --> 0:13:16.400 or conscious in some way, which they're not. When I 0:13:16.440 --> 0:13:19.040 say prompted, I mean we don't have all of the 0:13:19.080 --> 0:13:21.800 information as to how these models were trained or prompted 0:13:21.800 --> 0:13:25.040 to get these answers. They can make them do them. 0:13:25.040 --> 0:13:27.960 They could create a model to do it. And oh anthropic, 0:13:28.000 --> 0:13:32.160 wouldn't lie, wouldn't they, Dario Amat They went on stage 0:13:32.200 --> 0:13:35.680 and said they'll be twenty percent unemployment from white collar labors, 0:13:35.720 --> 0:13:37.280 and he'll claim me he said it was the future. 0:13:37.280 --> 0:13:40.839 It's a fucking lie, absolutely ridiculous. And I believe that 0:13:40.880 --> 0:13:44.200 the generative AI market is a fifty billion dollar revenue industry, 0:13:44.200 --> 0:13:47.120 masquerading is a trillion dollar one, and the media is helping. 0:13:47.760 --> 0:13:50.760 And underneath this troubling fact there's another one that should 0:13:50.760 --> 0:13:54.439 give us pause. The AI trade really is not about AI. 0:13:54.720 --> 0:13:58.120 It's not about generative AI, it's not about autonomous agents, 0:13:58.160 --> 0:14:01.679 it's not about AGI. It's all about GPUs, and it's 0:14:01.679 --> 0:14:14.120 incredibly brittle as a result. Now, as I've explained at length, 0:14:14.200 --> 0:14:16.640 the AI trade is not one based on revenue user growth, 0:14:16.640 --> 0:14:20.240 the effixive tools or significance of any technological breakthrough. Stocks 0:14:20.280 --> 0:14:22.560 are not moving based on whether they're making money or AI, 0:14:22.640 --> 0:14:25.280 because if they were, they'd be moving downward. However, due 0:14:25.320 --> 0:14:27.480 to the ViBe's based nature of the AI trade, companies 0:14:27.480 --> 0:14:31.800 are benefiting from the press inexplicably crediting growth to AI 0:14:31.880 --> 0:14:34.520 with no proof that that's the case. Open AI is 0:14:34.560 --> 0:14:36.920 a terrible business, But the only business worse than open 0:14:36.920 --> 0:14:39.080 AI are the companies built on top of them. Large 0:14:39.120 --> 0:14:41.440 language models are too expensive to run and have limited 0:14:41.440 --> 0:14:44.640 abilities beyond the ones I'm named previously, and because everybody's 0:14:44.680 --> 0:14:47.320 running models that all on some level do similar things, 0:14:47.320 --> 0:14:50.080 it's very hard for people to build really innovative products 0:14:50.080 --> 0:14:55.680 on top of them. But ultimately, this entire trade, all 0:14:55.720 --> 0:14:59.880 of the AI trade, hinges on GPUs. Coreweave was initially 0:15:00.040 --> 0:15:03.200 funded by Nvidia. They're a company that does AI data centers. 0:15:03.240 --> 0:15:05.640 I don't want to talk about them, makes me angry, 0:15:06.000 --> 0:15:09.840 but their IPO was funded partially buy and Video sells 0:15:09.880 --> 0:15:13.000 the GPUs. Video is also one of their customers, and 0:15:13.080 --> 0:15:17.480 core Weave raises debt to buy GPUs by using the 0:15:17.520 --> 0:15:20.880 GPUs they've already bought as collateral, and so they get 0:15:20.920 --> 0:15:24.160 those loans, and then they use that money to build 0:15:24.240 --> 0:15:27.280 data centers and buy more GPUs to put in them, 0:15:27.400 --> 0:15:31.200 which they then use as leverage to Do you see 0:15:31.240 --> 0:15:34.160 the Do you see the problem? Do you see the 0:15:34.240 --> 0:15:37.200 issue with that? This isn't me being polemic or hysterical, 0:15:37.240 --> 0:15:38.920 by the way, this is quite literally what's happening and 0:15:38.960 --> 0:15:40.920 how core Weave operates. If you want alarm by that, 0:15:40.960 --> 0:15:43.760 I'm not sure what to tell you Elsewhere. Oracle is 0:15:43.800 --> 0:15:46.520 buying forty billion dollars in GPUs for the still uninformed 0:15:46.520 --> 0:15:49.440 Stargate Data Center project, and Meta is building a Manhattan 0:15:49.520 --> 0:15:51.840 sized data center to fill within video GPUs. If you 0:15:51.880 --> 0:15:54.200 believe that Meta will actually do that, which I do not. 0:15:55.080 --> 0:15:57.640 Open AI is also Microsoft's largest as your client and 0:15:57.720 --> 0:16:00.720 insanely risky proposition on multiple levels, not simply in the 0:16:00.720 --> 0:16:03.160 fact that it's serving the revenue at cost, but the 0:16:03.200 --> 0:16:06.320 Microsoft executives believed that open ai would fail in the 0:16:06.320 --> 0:16:08.560 long term when they invested in twenty twenty three. According 0:16:08.600 --> 0:16:12.440 to the information, Microsoft is in Videa's largest client for GPUs, 0:16:12.640 --> 0:16:15.360 meaning that any changes to Microsoft's future and interest in 0:16:15.400 --> 0:16:18.320 open AI, such as reducing its data center expansion, would 0:16:18.320 --> 0:16:21.320 eventually hit in Vidia's revenue. Why do you think deep 0:16:21.400 --> 0:16:23.640 seek shocked the market. It wasn't because of any clunky 0:16:23.680 --> 0:16:26.640 story around training techniques or costs. It was because it's 0:16:26.640 --> 0:16:28.720 said to the market that and Video might not sell 0:16:28.760 --> 0:16:34.280 more GPUs every single quarter in perpetuity. Microsoft, Meta, Google, Apple, Amazon, 0:16:34.320 --> 0:16:37.560 and Tesla aren't making much money from AI. In fact, 0:16:37.600 --> 0:16:40.120 they're losing billions of dollars on whatever revenues they do 0:16:40.200 --> 0:16:42.520 make from it. Their stock growth is not coming from 0:16:42.560 --> 0:16:45.360 actual revenue or actual products, but the vibes around being 0:16:45.400 --> 0:16:48.560 an AI company, which means absolute jack shit when you 0:16:48.600 --> 0:16:51.200 don't have to users, finances, or products to back them up. 0:16:52.000 --> 0:16:54.800 So really everything on the AI trade comes down to 0:16:54.840 --> 0:16:57.560 in Video's ability to sell GPUs and more of them 0:16:57.560 --> 0:17:00.160 each quarter. And this industry, if we're really honest, is 0:17:00.160 --> 0:17:02.560 at a point where it really only exists to do 0:17:02.680 --> 0:17:06.440 so genera. If AI products do not provide significant revenue growth, 0:17:06.440 --> 0:17:08.480 its products are not useful in the way that unlock 0:17:08.520 --> 0:17:11.879 significant business value, and the products that have some adoption 0:17:12.000 --> 0:17:17.399 run at such a grotesque loss. It's really sad. But 0:17:17.480 --> 0:17:19.399 I realize I've thrown a lot at you. And for 0:17:19.440 --> 0:17:21.480 the second time this year, I've recorded the three part 0:17:21.560 --> 0:17:24.680 podcast series. And what's great is you may think maybe 0:17:24.680 --> 0:17:27.640 I took a break. Maybe no. I recorded these bad 0:17:27.680 --> 0:17:29.960 boys back to back to back. That's why I sound 0:17:30.040 --> 0:17:32.280 so tired and why my voice kind of I think 0:17:32.320 --> 0:17:36.000 smooths out. I like doing this, but I needed to 0:17:36.040 --> 0:17:37.719 say all of this. I really did because I am 0:17:37.800 --> 0:17:42.080 genuinely really worried. We're in a bubble. If you do 0:17:42.160 --> 0:17:44.320 not think we're in a bubble, you are not looking outside. 0:17:44.440 --> 0:17:48.879 Apollo Global chief economist Torsten slock Killer name said it 0:17:48.960 --> 0:17:51.879 last week in a research note published July sixteenth. Okay, 0:17:52.000 --> 0:17:54.520 what he said was much worse. The difference between the 0:17:54.560 --> 0:17:56.639 IT bubble in the nineteen nineties and the AI bubble 0:17:56.640 --> 0:17:58.240 today is that the top ten companies in the S 0:17:58.280 --> 0:17:59.879 and p today are more overvalued than they were in 0:17:59.880 --> 0:18:02.680 the nineteen nineties. Slock wrote in a recent research note 0:18:02.720 --> 0:18:05.200 that was widely shared across social media. In financial circles, 0:18:06.280 --> 0:18:08.840 we're in a bubble. We're in a bubble general. If 0:18:08.840 --> 0:18:11.200 AI does not do the things that it's being sold 0:18:11.240 --> 0:18:13.520 as doing, and the things it can do aren't the 0:18:13.640 --> 0:18:16.000 kind of things that create business returns, automate labor are 0:18:16.000 --> 0:18:17.840 really do much more than one extension of a cloud 0:18:17.880 --> 0:18:20.800 service platform. The money isn't there, the users aren't there. 0:18:20.880 --> 0:18:22.959 Every company seems to lose money, and some companies lose 0:18:23.040 --> 0:18:25.320 so much money it's impossible to tell how they'll survive. 0:18:25.840 --> 0:18:29.800 Worse Still, this bubble is entirely symbolic the bailouts. And 0:18:29.840 --> 0:18:32.240 I know I get a lot of emails about this, 0:18:32.320 --> 0:18:35.399 and I'm really going to address this issue right now. 0:18:35.640 --> 0:18:37.840 The bailouts of the Great Financial Crisis were focused on 0:18:37.880 --> 0:18:39.720 banks and funds that have failed because they ran out 0:18:39.720 --> 0:18:42.359 of money, and the top initiative existed to plug holes 0:18:42.520 --> 0:18:47.040 with low interest loans. There are no few holes to 0:18:47.040 --> 0:18:50.480 plug here, because even if OpenAI and anthropics somehow became 0:18:50.480 --> 0:18:53.320 eternal money burners, the AI trade exists based on the 0:18:53.359 --> 0:18:57.239 continued and continually increasing sale and use of GPUs. There 0:18:57.240 --> 0:18:59.639 are limited amounts of capital, but also limited amounts of 0:18:59.720 --> 0:19:02.040 data is to actually put GPUs, and on top of that, 0:19:02.280 --> 0:19:04.040 at some point growth will slow at one of the 0:19:04.080 --> 0:19:07.040 magnificent seven, at which cost time costs will have to 0:19:07.080 --> 0:19:09.080 come down from things that lose them tons of money, 0:19:09.080 --> 0:19:11.600 such as generative AI. I really want to be clear, 0:19:11.640 --> 0:19:15.200 though a lot of people say, oh, it's really kind 0:19:15.240 --> 0:19:18.399 of lazy doomerism. I don't appreciate it, because, like, if 0:19:18.440 --> 0:19:20.320 you're going to be worried about something, be worried about 0:19:20.359 --> 0:19:23.080 something that's real, you're saying, oh, Trump will just bail 0:19:23.080 --> 0:19:25.440 them out. Oh, they'll get bailed out. Who will get 0:19:25.440 --> 0:19:28.879 bailed out? For this fast to continue? In video? Is 0:19:28.920 --> 0:19:31.359 thirty nine point one billion dollars of data cent revenue 0:19:31.359 --> 0:19:32.840 in the last quarter. That means the next one needs 0:19:32.880 --> 0:19:34.399 to be forty four and one after that needs to 0:19:34.440 --> 0:19:38.320 be like fifty sixty eighty one hundred. This needs to 0:19:38.359 --> 0:19:42.119 continue forever. Is the government going to nationalize in video? 0:19:42.280 --> 0:19:42.359 Is? 0:19:42.760 --> 0:19:44.080 Well? I guess that would take them off the market, 0:19:44.119 --> 0:19:46.760 wouldn't it is they're just going to feed in video 0:19:46.840 --> 0:19:49.600